bottom up parsing - ksu faculty

Bottom Up Parsing

• More general than deterministic top-down parsing– Just as efficient– Uses the same ideas

• The method used by most compiler generation tools

• +ve: do not need left factored grammar• For example the following grammar is OK

E→T+E|TT → int*T | int |(E)

• A grammar is OK provided that it is unambiguous

What is Bottom-Up Parsing?

• Idea: Apply productions in reverse to convert the user's program to the start symbol.

• A left-to-right, bottom-up parse is a rightmost derivation traced in reverse (as we will see).

Important Fact #1 about bottom-

up parsing:

A bottom-up parser traces a

rightmost derivation in reverse

Shift Reduce Parsing

• The main strategy used by bottom up parsers• Recall that a bottom-up parser traces a rightmost derivation in

reverse• An important consequence

– Let αβω be a step of a bottom-up parse– Assume the next reduction is by X→ β

– Then ω is a string of terminals, otherwise the reduction we just did was not for the rightmost terminal.

• The general idea: split string into two substrings– Right substring is yet unexamined by parsing– The Left substring has terminals and non-terminals– The left substring is our work area (where we should search for

handles)– The dividing point is marked by a |

Two Main kinds of actions

1. Shift: Move | one place to the right• i.e., shifts a terminal to the left string

• e.g. ABC|xyz ABCx|yz

2. Reduce: apply an inverse production at the right end of the left string– If A→ xy is a production, then

Cbxy | ijk CbA| ijk

• When to shift and when to reduce is another story

Bottom-Up Parsing Using a Stack

• Left string can be implemented by a stack

– Shift

• pushes a terminal on the stack

– Reduce

• Pop symbols off of the stack (the right hand side of a production)

• Pushes a non-terminal on the stack (the left hand side of a production)

Conflicts

• In a given state, more than one action (shift or reduce) may lead to a parse tree

• Two main kinds of conflicts:– If it is legal to shift or reduce, there is a shift-reduce

conflict• Not very good, but it is easy to rewrite the grammar to

remove it.

– If it is legal to reduce by two different productions, there is a reduce-reduce conflict.• This is bad because it indicates that something is wrong with

the grammar

Handles

• How do we decide when to shift or reduce?

• The leftmost reduction is not always the best thing to do

• 2 Examples

Another Example

• Example grammar:

– E→ T+E | T

– T → int * T | int | (E)

– Consider step int | * int + int

• We could reduce by T → int giving T| *int+int

• A fatal mistake– Since no production can handle T*

– There would be no way to reduce to the start symbol E

Handles

• A handle is a reduction that allows further reductions back to the start symbol

• S→* αXω →αβω, then αβ is the handle simply because it is not a mistake it allowed us to go back to the start symbol.

• The handle of a parse tree T is the leftmost complete cluster of leaf nodes.

• A left-to-right, bottom-up parse works by

• iteratively searching for a handle, then

• reducing the handle.

Finding Handles

• Where do we look for handles?

– Where in the string might the handle be?

• How do we search for possible handles?

– Once we know where to search, how do we identify candidate handles?

– What algorithm do we use to try to discover a handle?

• How do we recognize handles?

– Once we've found a candidate handle, how do we confirm that it is correct (i.e., the handle?)

Recognizing Handles

• There are no known efficient algorithms to recognize handles

• But, there are good heuristics for guessing handles

• On some CFGs, the heuristics always guess correctly

• It is not obvious how to detect handles

• At each step the parser sees only the stack, not the entire input;

• It sees α where α is a viable prefix if there is an ω such that α| ω is a state of a shift-reduce parser.

• Recall that α is on the stack while ω is the unseen input.

Viable Prefix

• A viable prefix because is a prefix of the handle

• In other words: it does not extend past the right end of the handle

• As long as a parser has viable prefixes on the

stack no parsing error has been detected.

Important Fact

• For any grammar, the set of viable prefixes is a regular language.

• Therefore they can be recognized by a finite automata

• The basis for many compiler generation tools.

• We will see how to construct such a FA.

• But first we need a few more definitions

An item

• An item is a production with a “.” somewhere on the rhs.

• The items for T → (E) are

T → .(E)

T → (.E)

T → (E.)

T → (E).

• The only item for X → ξ is X → .

• Items are often called “LR(0) items”

The problem in recognizing viable prefixes

• is that the stack has only bits and pieces of the rhs of productions

– If it had a complete rhs, we could reduce

• In any successful parse theses bits and pieces are always prefixes of rhs of productions.

• Consider the input (int)

E → T + E | T

T → int*T | int | (E)

– Then (E|) is a valid state of a shift-reduce parse

– (E is a prefix of the rhs of T → (E)

• Notice that it will be reduced after the next shift

– Item T → (E.) says that so far we have seen (E of this production and hope to see )

• i.e. no parsing errors so far

The structure of the stack

• The stack does not contain an arbitrary string of symbols.

• The stack may have many prefixes or rhs’s

• Prefix1 Prefix2 … Prefixn-1 Prefixn

• Let Prefixi be a prefix of rhs of Xi → αi

– Prefixi will eventually reduce to Xi

– The missing part of αi-1 starts with Xi

– i.e., there is an Xi-1 → Prefixi-1 Xi β for some β

– Recursively, Prefixk+1 … Prefixn eventually reduces to the missing part of αk

An Example• Consider the grammar

E → T + E | T


• And the string (int * int)

• (int * | int) is a state of a shift-reduce parse

• The stack contents from bottom-to-top is

“(” which is a prefix of the rhs of T → (E)

“ξ” which is prefix of the rhs of E → T

“int *” which is a prefix of the rhs of T → int * T

• The stack of itemsT → (.E)

E → .T

T → int * .T

• Which says

– We have seen “(“ of T → (E)

– We have seen ξ of E → T

– We have seen int* of T → int * T

• To recognize viable prefixes, we must

– Recognize a sequence of partial rhs’s of productions, where

– Each partial rhs can eventually reduce to part of the missing suffix of its predecessor

An Algorithm for recognizing Viable prefixes

• Recall that the set of viable prefixes are regular, so what we are going to do is to construct a NFA that recognizes them.

• The input of the NFA is the stack.

– It will be read bottom-up

• The output is

– yes if it is a viable prefix and

– no if it is not.

• The states of the NFA are the items of the grammar.

Algorithm: 1. Add a dummy production S’ → S to G

– this makes S’ the new start symbol and

– makes sure that there is one production for the new start symbol

2. The NFA states are the items of G– Including the extra production

3. For item E → α.Xβ add transition– (i.e. so far we have seen α on the stack)

– So if x is the next symbol on the stack (above α) then we can make this transition

– E → α.Xβx E → αX.β where X a terminal or non terminal (i.e., a move the NFA can make)

4. For item E → α.Xβ and for every production X →ϒ– where X is a non-terminal

– and what is on the stack can eventually be reduced to x

– So we can make the transition

– E → α.Xβ ξ X → .ϒ

5. Every state is an accepting state

6. Start State is S’ → .S

ExampleS’ → E (the extra production)

E → T + E | T


The start state is the extra production

What transitions can we make depends on what can be on the stack: There can be an E on the stack or something derived from E so we need three transitions

Notice the dot on the left of T, indicating that we are hoping to see T on the stack next.

Notice the dot on the left of T, indicating that we are hoping to see T on the stack next.

The transitions for E→.T

Notice the dot on the right indicates that we are ready to perform a reduction.

If we do not see a T, we may see something that can be derived from T

The transitions for E → .T+E

The transitions for T → .(E):One possible transition that is if we see “(“ on the

stack.One possible transition that is if we see “(“ on the stack.

The transition for T → (.E)

The transition of T → (E.)

We can make a reduction

The transition for E → T.+E

Transitions for E → T+.E

Transitions for T → .int

Transitions for T → .int * T

Transitions for T → int. * T

Transitions for T→ int *.T

An Equivalent DFA

the start state

•Notice: that each item is state and

•the NFA can be in any of these states

• The states of the DFA are “canonical collections of items”

• Or “canonical collections of LR(0) items”

• Item X→ β.ϒ is valid for a viable prefix αβ if

S’ →* αXω → αβϒω

by a rightmost derivation

• After parsing αβ, the valid items are the possible tops of the stack of items.

Valid Items

• An item is often valid for many prefixes

• Example: The item T → (.E) is valid for prefixes(

((

(((

((((

(((((

…

SLR Parsing Algorithm: Simple LR parsing

• LR(0) parsing: Assume

– Stack contains α

– Next input is t

– DFA on input α terminates in state s

• Reduce by X → β if

– S contains item X → β. (i.e. we have seen a complete rhs)

• Shift if

– S contains item X → β.tω

– i.e. s has a transition labeled t

2 kinds of problems

• LR(0) may not be able to decide what to do in two situations

• LR(0) has a reduce/reduce conflict if:

– Any state has two reduce items:

– X → β. and Y → ω. (two possible reduce actions)

• LR(0) has a shift/reduce conflict if:

– Any state has a reduce item and a shift item:

– X → β. and Y → ω.tδ• (i.e. a reduce is possible and a shift is also possible)

A shift reduce conflictA shift reduce

conflict

SLR Parsing

• SLR = “Simple LR”

• SLR improves on LR(0) by adding shift/reduce heuristics that

• will help us determine when to reduce and when to shift

– Fewer states have conflicts

SLR Parsing

• Assumptions

– Stack contains α

– Next input is t

– DFA on input α terminates in state s.

• Reduce by X → β if

– S contains X → β.

– And

– t ϵ Follow(X) (where t is the next input)

• Shift if

– s contains item X → β.tω

• If we still have conflicts after applying these rules, then the grammar is not SLR.

• The rules are heuristics for detecting handles

– The SLR grammars are those where the heuristics detect exactly the handles.

Parsing ExampleWe will reduce if the next input is in the Follow(E) which is {$,)}and shift if the next input is +

We will reduce if the next input is in the Follow(T) which is {$,),+}and shift if the next input is *

• Notice that all conflicts are resolved in the above grammar so it is an SLR grammar.

• But many grammars are not SLR

– These include all ambiguous grammar.

• We can parse more grammars by using precedence declarations

– Instructions for resolving conflicts

• Consider the grammar

• E→ E+E | E*E | (E) | int

• The DFA for this grammar contains a state with the following items:

– E→ E*E. And E→ E.+E

– Shift/reduce conflict.

• Declaring that * has higher precedence than + resolves this conflict in favor of reducing.

• So we will not do the shift.

SLR Parsing algorithm• Let M be a DFA for viable prefixes of G

• Let |x1…Xn$ be initial configuration (the stack is empty)

• Repeat until configuration is S|$ (i.e. until all input has been consumed)

– Let α|ω be current configuration

– Run M on current stack α

– If M rejects α, report parsing error

• Stack α is not a viable prefix

– If M accepts α and ends in a state with items I, let a be next input

• Shift if X→β.aϒ ϵ I

• Reduce if X→β. ϵ I and a ϵ Follow(X)

• Report parsing error if neither applies

A Parsing Example: int * int$

bottom up parsing - ksu faculty

Documents